AN ABSTRACT OF THE THESIS OF

advertisement
AN ABSTRACT OF THE THESIS OF
Tyler G. Mintkeski for the degree of Master of Science in Environmental
Science presented on June 19, 2006.
Title: The Utility of a Decision-Support Model to Assess Watershed
Condition for Salmon Recovery
Abstract approved:
Redacted for Privacy
Many contemporary fisheries and wildlife issues are complex,
messy, and divisive. Most share a set of common characteristics
including a lack of comprehensive scientific information, a limited
understanding of biological processes, a scarcity of agency staff,
time, money, and a tendency for differences over policy preferences to
end up as debat
over scientific information. When pressured to
provide policy relevant science to decision makers, agency scientists
are often left with no choice but to rely on some form of expert
opinion.
Information based on expert opinion may be valuable, but to
be most useful in decision making, it must be perceived as being
accurate, transparent, and calibrated by some measure of uncertainty.
Formal methods of eliciting and using expert opinion are becoming more
common in fisheries and wildlife management. Decision-support models
are one such method but are still fairly new and untested for fish and
wildlife problems. Using Oregon's Coastal Coho Salmon
kisutch
Oncorhynchus
Evolutionarily Significant Unit (ESU) as a case study, the
usefulness of Ecosystem Management Decision Support (EMDS) to assess
watershed condition for coho salmon recovery was examined. To create
the model, expert opinion was elicited using a formal Delphi process.
We found that the Delphi technique is relatively inefficient and
impractical for eliciting expert opinion on such topics as complex as
watershed condition. We also identified ways that normative science can
influence the consensus building process. Once our decision-support
model was constructed we determined that it was not particularly useful
for assessing watershed condition for coho salmon at the population
level due to the lack of data. Data for all of the road parameters in
the knowledge base were either nonexistent, or not available because
they are privately owned or require extensive GIS analysis. In this
case study we evaluated the tradeoffs of these formal methods:
improving credibility and transparency came at the cost of time and
procedural efficiency. Formal methods of eliciting and applying expert
opinion for assessments are no panacea. Managers and decision makers
will need to weigh these pros and cons on a case by case basis when
contemplating whether these tools will add appreciable value to their
assessments.
©Copyright by Tyler G. Mintkesk±
June 19, 2006
All Rights Reserved
The Utility of a Decision-Support Model to Assess Watershed
Condition for Salmon Recovery
by
Tyler G. Mintkeski
A THESIS
submitted to
Oregon State University
in partial fulfillment of
the requirements for the
degree of
Master of Science
Presented June 19, 2006
Commencement June 2007
Master of Science thesis of Tyler G. Mintkeski presented on June 19,
2006.
APPROVED:
Redacted for Privacy
Major Professor, representing Environmen/al Science
Redacted for Privacy
Dire
or of th
Envirbnmental Siences Graduate Program
Redacted for Privacy
Dean of th; Grduate School
I understand that my thesis will become part of the permanent
collection of Oregon State University libraries. My signature below
authorizes release of my thesis to any reader upon request.
Redacted for Privacy
Tyler3. Mintkeski, Author
ACKNOWLEDGEMENTS
To Bob Lackey, thank you for your patience and guidance.
To Grandfather, thank you for your financial support. Without you my
education would not have been possible.
To Maureen, thank you for your patience and your willingness to
compromise.
To Dad, thank you for your encouragement.
To Mom, thank you for pushing.
To my committee, expert panelists, and all those that offered advice
and friendship during this process, thank you for your inspiration and
passion for salmon and the watersheds in which they live.
TABLE OF CONTENTS
Page
INTRODUCTION
The Role of Expert Judgment in Salmon Recovery
CASE STUDY
1
1
9
Challenges of Assessing Watershed Condition
9
Ecosystem Management Decision Support (EMDS)
12
Oregon Coastal Coho Salmon ESU
17
METHODS
21
RESULTS
25
Delphi Process
25
Coho Decision-Support Model
26
DISCUSSION
34
Delphi Process
34
Utility of ENDS
40
CONCLUSION
46
BIBLIOGRAPHy
50
APPENDICEs
55
Appendix 1. Ecosystem Management Decision Support
56
Appendix 2. Example questionnaire provided to
expert panelists
61
LIST OF FIGURES
Figure
Page
The Oregon Coastal Coho ESU (Evolutionarily
Significant Unit)
8
Theoretical diagram of relationships between
watershed controls and physical processes on
habitat characteristics, and between habitat
and salmon fitness and survival
10
AREMP's decision-support model for assessing
watershed condition in the Oregon/Washington coast
The coho decision-support model
for the EMDS system built by experts in a Delphi
process to evaluate the ecological condition of
watersheds for the Oregon Coastal Coho ESU
.. 14
27
LIST OF TABLES
Table
Page
Coho decision-support model parameters, definitions
evaluations curves, evaluation criteria, data
sources and rationale
29
Parameters that failed at least one of the
criteria to be included in the model
33
Parameter categories and their suitability to make
assessments at 3 levels: ESU, monitoring area, and
population
41
LIST OF APPENDIX FIGURES
Figure
Page
Conceptual diagram of an EMDS decision-support
Model to assess watershed condition
57
1.2
Example evaluation curve
58
1.3
Example map of EMDS model evaluation results
60
2.1
Simplified diagram of the structure and function
of a decision-support model for the ENDS system
62
AREMP's decision-support model for assessing
watershed condition in the Oregon/Washington coast
65
1.1
2.2
The Utility of a Decision-Support Model to Assess
Watershed Condition for Salmon Recovery
INTRODUCTION
The Role of Expert Judgment in Salmon Recovery
Recovering US Sndangered Species Act (ESA) and Canadian Species
at Risk Act (SARA) listed stocks of anadromous salmon
Oncorhychus
spp. is a central focus of natural resource management agencies in
the Pacific Northwest of North America. Beyond the legal mandate, the
economic, social, and ecological importance of salmon in this region
is broadly accepted by the public, but specific recovery strategies
are controversial and recovery remains an illusive goal. After years
of analysis and the expenditure of billions of dollars, salmon runs
are still greatly reduced compared to pre-1850 levels (Meengs and
Lackey 2005)
Salmon recovery is an example of a contemporary natural
resource problem commonly known as a "wicked problem" (Rauscher
1999) .
Their shared characteristics include: inherent ecological
complexity, fragmented or competing interest groups and stakeholders,
a general lack of data, incomplete scientific understanding of the
problem and the consequences of management actions, large but unknown
costs, lack of organized approaches, and high uncertainty.
Additionally, the exact problem may be difficult to define and the
solutions obscured because of multiple interrelated, compounding
factors (e.g. changing political goals, dueling scientists, and the
2
continued use of ineffective approaches to solve these problems)
Often, these challenges are exacerbated by a lack of financial
resources and time constraints.
Given that it has the characteristics of a wicked problem,
effective salmon recovery requires the collaboration and
communication between scientists and decision-makers. In this
relationship, decision-makers are under pressure to implement polices
that achieve their organization's specific legal or policy mandates
whereas scientists are expected to inform management decisions by
providing the "best available scientific information" that is timely,
relevant, and meaningful. Bisbal (2002, 2006) suggests that, although
popular in use and even required by the ESA to guide salmon
conservation, the definition of the "best available science" is
vague. There is no consensus on what counts as the "best available
science", how to differentiate between the best science and the rest,
or how to determine the amount of science available (Bisbal 2002,
Bogert 1994).
When confronted by the wicked characteristics of the
problem of salmon recovery, particularly incomplete data coupled with
complexity and urgency, the "best available science" is often expert
judgment.
Expert judgment is a valuable, even essential, source of
information for the management and recovery of salmon (Magnuson et
al. 1995) . From national scientific panels to individual fisheries
biologists, scientists interpret information and provide judgment at
many levels that influence salmon management decisions. It is assumed
3
that experts have the ability to make judgments and decisions based
on incomplete data and imprecise understanding (Regan et al. 2004,
Johnson and Gillingham 2004). Experts' additional challenge is to
synthesize expert opinion in a way that is easy to understand and
useful for decision making. In contrast to inferences based solely on
empirical data, experts can provide a synthesis perspective gathered
from a coirtiDination of experience, personal observation, and published
data (Johnson and Gillingham 2004). Also, expert judgment can
substitute for empirical data that is cost prohibitive and time
consuming to collect (Rushton et al. 2004, Johnson and Gillingham
2004)
The use of expert judgment deserves caution based on its
subjective nature. An expert may have a sound technical understanding
of a subject, but there is no assurance that an expert's judgment
process will follow the rules of rational thought1 or that the
judgment will amount to useful information. Expert-derived
information is susceptible to being undermined by personal values,
perceptions of risk, and cognitive heuristics2 (Regan et al. 2004)
Also, an expert's reasoning is not transparent. It is impossible to
'According to Cleaves (1994) rationality in judgment means "the
person who estimates, values, or chooses thoroughly uses available
information and is aware of alternatives and their implications and
that the judgment is coherent, consistent with similar judgments, in
agreement with general laws or probability, and understood by users.
2Heuristics are psychological rules which have been proposed to
explain how people make decisions, come to judgments and solve
problems, typically when facing complex problems or incomplete
information. These rules work well under most circumstances, but in
certain cases lead to systematic cognitive biases (Tversky and
Kahneman 1982).
4
understand an expert's logic or hidden reasoning underlying his or
her assumptions (Regan et al 2004)
.
Therefore, expert based
assessments and decisions are difficult to repeat. Lack of
repeatability can potentially weaken the credibility of decisions and
assessments based on expert judgment by hiding inconsistencies,
values, illogical assumptions, or "normative" science F"science which
presupposes a particular policy position or perspective")
2004) .
(Lackey
As a result, expert judgment that is presumed to be biased
may be not be accepted for useeven if it is not biased. Scientists
perceived as biased have lost their credibility, and policymakers
will likely ignore any scientific information they provide (Rykiel
2001) .
The worst case is when management decisions are made based on
biased or misleading information that unexpectedly and irretrievably
damages the resource or result in undesirable consequences (Johnson
and Gillingham 2004).
Like scientific information, expert judgments are most useful
if they are calibrated by a measure of uncertainty, communicated
intelligibly and meaningfully, and most importantly, if they are
perceived as credible (Ellison 1996) . The credibility of expert
judgment is dependent on its transparency, repeatability, and
defensibility, qualities affected in part by how expert judgment is
elicited. Methods of eliciting expert judgment may be formal or
informal. Formal methods for elicitation explicitly document
information and sources, use consensus building, and demonstrate
transparently logic paths and assumptions. One formal method of
expert solicitation is the Delphi technique in which judgment is
5
solicited anonymously from an expert panel, summarized, and returned
to the each expert for reconsideration (Crance 1987). The process is
repeated until consensus is reached.
The Delphi technique is just one formal method to elicit and
integrate expert opinion. Alternatives methods to the Delphi
technique include one-shot group averages, group discussions, and the
Nominal Group technique. One-shot group averages provide a single
chance for expert panelists to provide their input before averaging
the group's answers. Its advantage is speed. In general, one-shot
group averages lack enough time for participants to give their most
thoughtful input and do not allow participants to hear each other's
ideas that could influence and help refine their own ideas. Group
discussions between experts are unrestricted during a face-to-face
meeting. They are administratively difficult to coordinate and
expensive to bring all participants together. Group discussions are
susceptible to group dynamics (e.g. dominant individuals control the
decision process) and can obscure objective information (Clayton
1997). The Nominal Group Technique uses the Delphi process in a face-
to-face setting without the anonymity of expert input (participants
read their answers aloud to the group for feedback)
1975) .
(Delbecq et al.
In addition to sharing the same cost and administrative
problems as the basic group discussion method, the lack of anonymity
may influence how individuals participate. The advantages of the
Nominal Group technique are structured input and summarized feedback.
Informal methods to elicit expert judgment have no formal rules for
6
how information is gathered or combined and may be as simple as
asking a single expert a few casual questions.
Once elicited, expert judgment is applied to the decision
making process. Like elicitation, these methods of decision making
can also be informal or formal. Formal methods are defined as models
that demonstrate more transparently how a specific conclusion is
reached. As an example, decision support models are computer models
that capture logical evaluation procedures for consistent application
to a decision process (Gallo et al. 2005). They allow the user to
observe how model parameters are related and where data synthesis
occurs in the evaluation process. When discrepancies in model outputs
occur, the causes can be determined. Informal methods (e.g. black box
models) provide little ability judge the quality or validity of
resulting assessments because the user cannot evaluate the inner
workings of the model. Consequently, formal methods of acquiring and
applying expert judgment are most credible in decision making.
Neither informal or formal methods for eliciting or applying
expert judgment is inherently the best choice. Both methods have pros
and cons and are suitable for different circumstances and situations.
Essential considerations include time constraints, data availability,
resources, complexity of goals, the number of experts, whether or not
consensus is needed, and then subsequently, how to build consensus.
Formal methods require more time, money, planning, technical models,
explicit procedures, and depending on the problem, a large number of
experts. Informal methods are time-efficient, have no constraints or
7
formal rules, and may be as simple making a few phone calls to a
single expert for advice.
The purpose of this paper is to evaluate, through a case study,
the utility of foLlual methods to both elicit and apply expert
judgment to the salmon recovery. Using Oregon's Coastal Coho Salmon
Oncorhynchus kisutch
Evolutionarily Significant Unit (ESU) as a case
study (Figure 1), the usefulness of a decision-support model to
assess watershed condition is evaluated. We ask: do using formal
elicitation and application methods add appreciable value to the
assessment? To do this an expert panel was assembled and instructed
to build a decision-support model for the Ecosystem Management
Decision Support (EMDS) system designed specifically to assess the
ecological condition of watersheds for coho salmon within the ESU.
The purpose of the model is to provide a transparent, objective,
expert-based tool for the comprehensive and consistent assessment of
each of the 21 major basins in the ESU (Figure 1)
.
Our interests were
as much about the process of formally eliciting expert judgment as
they were about the product- the coho decision-support model. The
three objectives were to determine:
(1) if eliciting expert judgment
using a formal method (Delphi technique) added utility to the
process;
(2)
if experts could agree on a model of watershed condition
for coho within a realistic time frame; and (3) if the coho decisionsupport model could be built and function with existing data.
8
Necanicum River
Nehalem River
Tillamook Bay
E North Coast
Nestucca River
Mid-Coast
U
Umpqua
Salmon River
M,d-South Coast
Siletz River
Beaver Creek
50km
Yaquina River
Col
Alsea River
co
I uu
Siuslaw River
4
Siltcoos River (Lake)
Tahkenitch Lake
Lower Umpqua River
Middle Umpqua River
Tenmile Lake
North Umpqua River
Coos Bay-Coquille River-----
Floras Creek
Sixes River-
South Umpqua River
Figure 1. Map of the Oregon Coastal Coho ESU (Evolutionarily
Significant Unit)
This map shows 3 spatial resolutions: (1) ESU. The
Oregon Coastal Coho ESU extends from the Columbia River south to Cape
Blanco; (2) Monitoring Area (colored areas)
The ESU is divided into
4 distinct monitoring areas for data collection; and (3) Population.
The ESU contains 21 independent populations of coho, which correspond
to the major river or lake basins along the coast.
.
9
2. CASE STUDY
Challenges of Assessing Watershed Condition
Deciding if, where, and when to restore habitat to help recover
salmon is one example of a wicked problem that is dependent in part
on expert judgment. Habitat restoration is a controversial issue
because of high costs and uncertain benefits to salmon recovery,
disagreements over how and where funds are allocated, and potential
for social dislocation (e.g. restricted property rights and land
uses)
Watershed assessment is the first step in the process of
habitat restoration and helps determine the recovery potential of
watersheds. Assessment is a difficult task because the factors that
comprise watershed condition are interrelated, with compounding and
potentially counterintuitive influences on salmon. The ecological
condition of a watershed is a function of interacting
geomorphological surfaces, physical processes and biological
communities further influenced by natural and human disturbances
(Stanford and Ward 1992, Pess et al. 2002)
(Figure 2)
Beyond the inherent complexity of watersheds, successful
assessments of watershed condition must overcome many technical
challenges (Reeves et al. 2004, Dai et al. 2004), including:
(1)
making consistent evaluations across watersheds with differing
spatial extents;
(2) aggregating multiple, diverse, and potentially
10
confusing variables;
(3) characterizing and quantifying
relationships.
Vegetation
0
Geology
Climate
0
U
Sediment
Supply
Orga nic
Hydrologic
Regime
Matter
Inputs
Physical Habitat
Characteristics
Nutrient!
Chemical
Inputs
Light/He at
Inputs
Water Quality and
Primary Prod uction
Salmon Fitness
and Survival
Figure 2. Relationships between watershed controls and physical
processes on habitat characteristics, and between habitat and salmon
fitness and survival. The black boxes indicate controls not affected
by anthropogenic disturbances. Adapted from Pess et al. (2003)
among processes and variables (e.g. in-stream condition variables to
upland processes) ;
(4) combining the best available scientific data
with expert judgment objectively; and (5) communicating results that
are accurate, comprehensible, and relevant to diverse constituencies.
11
There is no universally accepted method for assessing watershed
condition. Agencies and organizations generally chose watershed
assessment methods for which they have sufficient existing data and
that help answer specific questions relevant to their policy goals.
Methods are often challenged by competing stakeholders and resource
users.
The most popular scientific approach to ecological condition
assessment relies heavily on statistical models.
The advantages of
using statistical models are empirical objectivity, rigor, and the
ability to test a hypothesis with little attachment to policy (ISAB
2003) .
Reeves et al.
(2004) contends that it is difficult to assess
watershed conditions using traditional statistical methods because
they are complex, require technical expertise, and their results can
be confusing to decision-makers. Ellison (1996) points out that few
decision-makers have the scientific training to interpret conclusions
based on "technical jargon". Also, assessing watershed condition
requires evaluating cumulative effects of many parameters. Generally,
statistical assessments evaluate multiple parameters independently,
but struggle to synthesize or combine them into an index (Reeves et
al. 2004, Dai et al. 2004)
.
In addition, the relationships among
watershed parameters and their influences on watershed condition are
not precisely understood, nor are they ever likely to be (Schmolt et
al. 2001).
Therefore, expressing these relationships credibly with
statistical models is difficult and often impossible.
12
Ecosystem Management Decision Support (EMDS)
It is becoming more common for watershed assessments to be
based largely on the judgments of expert biologists who arguably may
be better able to assess complex and information-poor problems (Lee
2000) .
One assessment method that formally incorporates expert
judgment with empirical data is knowledge-based decision-support
models (DSM). A knowledge-based DSM is a method of documenting a
formal, logical organization of information for evaluation and
interpretation (Reeves et al 2004)
.
It helps the user explicitly
organize and evaluate large quantities of diverse data and synthesize
these data into a single score based on user-defined rules. This
allows the evaluation process to be applied consistently and
transparently.
Ecosystem Management Decision Support (EMDS) system is a nonproprietary knowledge-based DSM developed cooperatively by the U.S.
Forest Service and the U.S. Environmental Protection Agency. As a
computer-modeling tool, EMDS integrates knowledge-based reasoning
(NetWeaver Logic Engine) into a GIS (ArcView) environment for
ecological landscape assessment (Dai et al. 2004)
.
It is not a
statistical or simulation model nor was it designed for prediction.
Rather it is an indicative model able to infer the condition of a
landscape based on the synthesis of data and model structure (Figure
3)
.
The creators of EMDS (Reynolds et al. 2002) assert two basic
reasons for using knowledge based reasoning for landscape
assessments:
(1) the components of the problem and their relations
13
are so inherently abstract that mathematical models are difficult or
impossible to formulate; or (2) a mathematical model is possible in
theory but current knowledge is lacking or too imprecise to create
one.
EMDs allows the user to define watershed condition, select the
parameters that characterize the watershed (physical and biological
indicators), define their evaluation criteria using evaluation curves
(Appendix 1 Figure 2), and determine the relations among all the
parameters in an explicit conceptual diagram (the knowledge-based
decision-support model)
(Figure 3) .
Once populated with spatially
explicit data the system evaluates and aggregates the data
hierarchically based on the user defined rules to determine overall
watershed condition. Model output comes in the form of tables,
graphs, and color-coded maps displaying watershed condition scores
(Appendix 1 Figure 3). One program in ENDS system can also rank the
influence of individual parameters on the overall model. Reynolds and
others explain the ENDS system (2000, 2002, 2003 and 2004)
.
Detailed
information is available on the ENDS website
(http: //www.institute.redlands .edu/emds/) and in Appendix 1.
ENDS has certain strengths, some of which are well documented,
that make it useful for assessing the ecological condition of
watersheds. It has an explicit and intuitive model structure that is
14
Vegetation
p
Condloori of vegetetion
in watershed
DrIvers
Conditon of variables
that drive or direcly
irriluence watershed
(A
condirion
Road.
Condiflon of roads in
watershed
WATERSHED
CONDITION
Cooddions are suitable
to susmat urablo runs of
whd cobs salmon
AVE
'S
Water QualIty
Evaluanon of water
quality parameters
/
Responses
Condloon of vanubies
that respond to dclsing
nfuoncns
Physical Condition
Enaluason of reach
physical coridihon
A
aUnibutes
sch Condition
Average of reach
condinon scores
Biological CondIb
AVE operators pass the average evaluation score
Evalueson of reach
biological condition
aItrIbutes
Figure 3. AREMP's knowledge-based decision-support model for
Washington/Oregon Coast Range Province used as a template for the
coho knowledge base. Modified from Gallo et al. (2005) . Examining the
model template from right to left show how individual parameters are
aggregated hierarchically into categories until all are combined in
an overall score of watershed condition. The general watershed
parameter categories are arranged vertically in 4 colored branches.
The top two are vegetation (green) and roads (brown), which are
combined to form the watershed condition "drivers" category. All of
the parameters in the drivers category directly influence watershed
condition. The bottom two categories are water quality (blue) and
reach condition (orange), which are combined to form the "responses"
category. All the parameters in this category respond to the drivers.
The reach condition score (orange) is the average condition score for
all the reaches in the watershed. Technically this is an independent
network, but it functions as part of the watershed condition
knowledge base and therefore the two are combined in the same diagram
for conceptual purposes. The reach condition score is an aggregate of
biological and physical conditions. The AREMP template incorporates
AVE operators at each junction (small boxes) where scores from
antecedent are averaged. None of the nodes are weighted.
easy to manipulate and communicate (Dai 2004, Bleier et al. 2003,
Reeves 2004, Gallo et al. 2005, Reynolds and Hessburg 2005) . The ENDS
map-based outputs of watershed evaluation scores are easy to
15
understand and communicate (Bleier et al. 2003, ISAB 2003)
.
The model
is able to combine empirical data from disparate sources (including
spatially explicit data) with expert judgment (ISAB 2003, Jensen
2000). Any part of the model (e.g. parameters, fuzzy curves, network
structure) can be easily
modified and updated as new knowledge and
information becomes available (NCWAP 2002, Reeves 2004, Dai et al.
2004, Gallo et al. 2005)
.
The ability of the modeling software to
"turn off" nodes allows EMDS to function with incomplete information
(Jensen et al. 2000, Pess et al. 2003). And, its ability to make
assessments as multiple spatial scales allows inclusion of multiple
agency jurisdictions and land ownerships (Pess et al. 2003, Dai et
al. 2004, Gallo et al. 2005)
EMDS also has weaknesses and concerns that limit its use:
(1)
it has weaker basis for scientific prediction than traditional
mathematical models validated with empirical data (ISAB 2003);
(2)
results of the model are qualitative and only indicate quality of
watershed condition (Dai et al. 2004); and (3) to be most useful
decision support models in general require communication between
analysts and decision makers about risk, probability, and uncertainty
which is difficult (ISAB 2003)
The use of EMDS for assessing ecological condition is recent.
Several studies have used it to assess watershed condition (Jensen et
al. 2000, Reynolds et al. 2000, Reynolds and Reeves 2003, Pess et al.
2003, Dai et al. 2004, Reynolds and Reeves 2004, Reynolds and
Hessburg 2005). In general these papers unveil EMDS as a new tool,
16
describe its abilities, and provide small case studies or examples of
its use to assess watershed condition. Reynolds and Reeves (2003 and
2004) also developed prototype models to evaluate habitat suitability
for salmon but were only used to demonstrate the abilities of EMDS.
Examples where ENDS has been applied to real ecological
problems are limited. Three interagency programs have adopted ENDS as
a tool to drive landscape assessment programs. The Rogue Basin
Restoration Technical Core Team (RBRTT) used EMDS to assess watershed
condition and identify restoration priorities in the Applegate River
sub-basin in Southwestern Oregon (RBRTCT 2004). California's North
Coast Watershed Program (NCWP) is using ENDS as an assessment tool in
a larger project to improve information and management of watershed
and fisheries conditions (NCWAP 2002)
.
EMDS was selected specifically
to help evaluate and synthesize information on watershed and stream
conditions important to salmon (Bleier et al. 2003)
.
NCWP experts
are still in the stages of refining the knowledge base for their
model.
The multi-federal agency Aquatic and Riparian Effectiveness
Monitoring Program (AREMP) is the largest ongoing effort using EMDS.
AREMP is characterizing the ecological condition of watersheds and
aquatic ecosystems for the area managed under the Northwest Forest
Plan (Reeves et al. 2004). The program's goal is to report on the
Northwest Forest Plan's effectiveness across the region, to track
trends in watershed condition over time, and to provide information
for determining causal relationships to help explain those trends.
17
ABNEP's models use monitoring data to estimate watershed condition by
aggregating upsiope, riparian, and in-channel indicators (Figure 3).
Prior to using EMDS, the U.S. Forest Service and the Bureau of Land
Management evaluated watershed variables independently and therefore,
could not easily determine ecosystem or watershed condition as a
whole (Reeves et al. 2004) . After review of other methods, ENDS was
chosen because of its ability to deal with uncertainty, incorporate
expert opinion, and guide monitoring efforts. It has been useful to
decision-makers in helping to identify changes in watershed condition
in the last 10 years.
Oregon Coastal Coho Salmon ESU
The Oregon Coastal Coho Salmon Evolutionary Significant Unit
(ESU)
is well-suited for testing the usefulness of EMDS to evaluate
watershed condition for salmon recovery decisions. The ESU includes
all naturally spawned populations of coho salmon in Oregon coastal
streams south of the Columbia River and north of Cape Blanco (Figure
1)
Pi ESU is defined legally as a distinctive group of Pacific
salmon in terms of their evolution and reproductive isolation from
other conspecific populations. The ESU contains 21 independent
populations of coho salmon, which correspond to the major river or
lake basins along the coast. Independent populations are those
thought to occur in basins with sufficient habitat in the past to
have persisted through several hundred years of normal variations in
marine and fresh water conditions (Nicholas et al. 2005)
18
Since 1997, The Oregon Coastal Coho EStJ has been the focus of
a collaborative conservation effort developed under a planning
framework called the Oregon Plan for Salmon and Watersheds (Oregon
Plan). The Oregon Plan, administered by the Oregon Watershed
Enhancement Board (OWEB), brings together various governmental
entities, interest groups, and stakeholders to implement salmon
restoration and conservation strategies.
In the spring of 2005, OWES completed a formal assessment of
the Oregon Coastal Coho ESU (hereafter referred to as the Coho
Assessment) to evaluate Oregon Plan conservation efforts targeting
the ESU and also to help inform the federal government's ESA listing
decision. Based in part on the conclusions of the assessment, NOAA
Fisheries decided to drop its proposal to re-list the ESU as
threatened in January of 2006. Whether listed under the ESA or not,
wild coho salmon populations in the ESTJ are well below historical
levels and so are of continuing policy concern (Meengs and Lackey
2005). Multiple stakeholders with largely mutually exclusive policy
goals continue to disagree over the listing status, the credibility
of science that support it, and the thoroughness of Oregon's Coho
Assessment in particular.3 In the midst of the controversy, Oregon
has an ongoing, long-term commitment to recover and conserve the ESU.
The recent Coho Assessment is a foundation for future conservation
Stakeholder comments about the Coho Assessment can be viewed on the
following webpage:
ftp: !/nrixnp.dfw. state.or.us/oregonplan/reports/Comments/
19
activities including recovery planning, monitoring, and more
effective investment of restoration funding
The Coho Assessment examined threats to the viability of the
ESU including: marine habitat, harvest, hatchery impacts, stream
complexity, fish passage, water quality, water quantity, and other
factors. Three spatial resolutions were evaluated: the entire ESU,
the monitoring area, and the area corresponding to independent
populations (Figure 1).
OWEB chose to focus the evaluations at the
population level because it is at this level that restoration and
management actions are targeted to improve the viability of the ESU
as a whole (Nicholas et al. 2005) . The Coho Assessment required
gathering a large, diverse suite of the best available scientific
data and information from state and federal natural resource agencies
[U.S. Forest Service (USFS), Bureau of Land Management (BLN), Oregon
Department of Environmental Quality (ODEQ), Oregon Department of Fish
& Wildlife (ODFW), Oregon Department of Forestry (ODF)], private
entities, and watershed councils. This was the first attempt to pull
together all the monitoring data collected for the ESU since the
start of the Oregon Plan in 1997. Oregon state agencies alone spent a
total of $15.9 million on coho related monitoring from 1997 to 2003,
not including monitoring by watershed councils, federal agencies, or
private landowners (OWEB 2005).
Using this information and expert judgment in multiple group
work sessions the Coho Assessment team identified the primary and
secondary risk factors for each of the 21 independent populations.
20
This process, lasting over a year, concluded that stream complexity
was the most common primary risk factor (13 of 21 populations) and
water quality was the most common secondary risk factor (15 of 21
populations)
(Nicholas et al. 2005) . How these conclusions were
reached is not transparent and may not be repeatable because the
informal process of building consensus of opinions is not explicitly
described in the Coho Assessment.
Considering these conclusions, the large suite of available
data, the reliance on expert judgment, and the goal of managing the
ESU with a watershed-scale approach (GNRO 2005), a formal assessment
of watershed condition for each of the 21 populations could be
valuable for the management and recovery of the ESU.
While watershed
assessments have been completed for over 97% of the ESU (Nicholas et
al. 2005), they have been completed by different organizations, using
various methods, at different scales, with different information and
are therefore difficult to compare.
We chose the Coastal Coho ESU case study because there is a
need and opportunity for an alternative watershed assessment method
that is transparent and repeatable, systematic, comparable,
comprehensive, and easy to understand. ENDS is one tool that could be
used to assess the coarse-scale status and trends of condition for
the 21 watersheds (corresponding to the independent populations),
thereby providing useful information for recovery activities. The
EMDS system could also provide an alternative method for determining
what specific factors are limiting to coho and thus provide a
21
starting point for estimating the recovery potential for each
watershed.
METHODS
An expert panel was assembled to build a knowledge-based DSM
for the EMDS system to assess the ecological condition of watersheds
for coho salmon within the Coastal Coho EStJ (hereafter coho decision-
support model). Ideal panel size varies in the literature from 5-25
experts (Clayton 1997, Crance 1987) . Potential experts were chosen
subjectively based on three criteria:
(1) high familiarity with coho
salmon biology and ecology on the Oregon coast;
(2) well established
in their careers; and (3) credible reputations for producing and
contributing sound science. Panelists were selected that were
perceived to not have a stake in the outcome of the project because
the model had the potential to influence future coho salmon
management decisions. We decided to exclude potential expert
panelists who had participated in the recent Coho Assessment. With
their intimate knowledge of the assessment's results and
controversies, we thought that they could potentially influence the
process and results. A total of 16 experts were identified and
solicited to participate but only 6 volunteered. Professionally, the
experts on the panel to build the coho decision-support model
represented state and federal agencies, non-profit organizations, and
private consulting firms specializing in coho salmon related,
research and management. All participants were ensured complete
confidentiality.
22
One panelist had participated in 2BEMP's process to build
decision-support models to evaluate watersheds in the region of the
Northwest Forest Plan. Although this expert had the advantage of
previous experience with decisions-support models, the expert was
allowed to participate because so few others volunteered.
Expert judgment was elicited from the panelists using the
Delphi technique (Crance 1987). Delphi is a formal, structured
process for soliciting and aggregating individual judgments about a
specific topic into group consensus using questionnaires. The Delphi
concept is based on the premises that opinions of experts are
justified as inputs to decision-making where absolute answers are
unknown; and a consensus of experts will provide a more accurate
response to a question than a single expert (Crance 1987)
.
The Delphi
process occurs via correspondence eliminating the need for
participants to travel and coordinate schedules. This also
facilitates the requirement that panel members participate
anonymously. Anonymity reduces disruptive social-emotional behaviors
that diminish focus on task-oriented activities and compromise panel
members' responses (e.g. pressure to conform)
(Clayton 1997).
The Delphi exercise to create the coho decision-support model
began with submitting a questionnaire to each of the expert panelists
(Appendix 2). Each panelist completed the questionnaire by answering
the questions, providing comments, and rating the confidence of their
answers. Throughout the exercise the panelists were asked to submit
23
written comments on their frustrations, concerns, and suggestions
about the model. When the questionnaires were returned, the moderator
summarized the comments, tabulated the answers and submitted a
follow-up questionnaire including the same set of questions, a
summary of the feedback from the previous round, and an updated model
diagram. The panelists then had the opportunity to re-answer the
questions after reviewing all of the other participants' anonymous
comments. All suggested changes and additions to the model had to be
approved by the majority of the expert panelists.
before the Delphi process began each panelist was provided
with:
(1) the panel's objective to create an ENDS model to assess the
ecological condition of watersheds for coho salmon within the Coastal
Coho EStJ;
(2) the definition of watershed condition4;
(3) background
EMDS information (Appendix 2); and (4) AREMP's skeleton knowledge
base for evaluating the ecological condition of Pacific Northwest
coastal watersheds as a template (Figure 3)
.
In an anticipated effort
to reduce the time to teach the experts about decision-support
models, AREMP's model was provided as an example of knowledge base
For the coho decision-support model a watershed in "good" condition
means that it has the ability to provide high-quality habitat for
coho salmon. In other words, watershed conditions are suitable to
sustain viable runs of wild coho salmon; where "viable" signifies
that coho salmon populations generally demonstrate sufficient
abundance, productivity, distribution, and diversity to be sustained
under the current conditions (Wicholas et al. 2005)
Therefore,
watershed processes (represented by parameters) are intact, and
functioning adequately to provide and maintain habitat for coho
salmon and other native species that compose the ecosystem. The
system must be able to recover to desired conditions when disturbed
by either natural events or anthropogenic activities (Reeves et al.
.
2004)
24
structure and logic. AREMP's model is the most comprehensive
knowledge-based decision-support model for watershed assessment and
has the longest record of use (Gallo et al. 2005)
.
Panelists were
also provided with a list of all the relevant watershed parameters
that were used in some form in Oregon's Coho Assessment, representing
data from state and federal agencies used to characterize instream,
riparian, upland and biological conditions (Appendix 2 Table 1).
The questionnaire charged the panelists with coming to
agreement on what parameters should be included in the model, the
structure of the model, and how the model should operate based on
available data, literature, and their own knowledge and opinions. The
first question was selecting parameters to use in the coho decision-
support model. Parameters were required to meet the following
criteria:
(1) they must act as surrogates or indicators of watershed
processes that influence coho; (2) data for the parameters must
currently exist (from field surveys or GIS) and be available to the
public; and (3) data for the parameters must exist for the entire
region encompassed by the coastal coho ESTJ. Once the parameters were
selected, the moderator assessed their ability to meet the criteria
and then constructed their evaluation curves (Appendix 1 Figure 2)
using existing data and literature. The second question was to
develop the coho decision-support model structure. The third question
was to decide how to aggregate parameters scores at each junction in
the model. The final question was to weight any parameters deemed
necessary.
25
As the moderator, I was in charge of all administrative tasks
including corresponding with the panelists (e.g. answering questions
and enforcing deadlines), creating questionnaires, and processing
panelist responses.
RESULTS
Delphi Process
The Delphi process to build the coho decision-support model
lasted 8 months and included 4 rounds of questionnaires. Even though
the questionnaire instructed experts to respond within 10 days, the
questionnaires were returned between 3 days and 7 weeks. The rate at
which questionnaires were summarized and returned in each subsequent
round was effectively controlled by the slowest participant. Crance
(1987) estimated that amount of time between mailing two consecutive
questionnaires in a Delphi exercise is 4-6 weeks. In this project,
the range was 4-8 weeks. In two instances a questionnaire from a
single participant was not returned and no explanation was given when
prompted requiring the process to continue on to the next round
without input from that panelist.
Two members on the expert panel dropped out at different times
during the process due to other obligations and priorities. After
soliciting various other potential expert panelists, only one vacancy
was filled for the last round in the process. There was no way to
measure the influence of the new expert panelist, but it was clear
26
from the quality and thoroughness of the comments that the panelists
was very engaged in the process and might have influenced the model
more if having participated from the beginning.
The amount and quality of feedback in written responses varied
considerably among participants. Some participants wrote extensively
while others replied with single word answers. Certain participants
consistently used the diagram of the model to draw their ideas and
make suggestions with pictures, whereas others relied exclusively on
written input. Overall, amount of information returned from 4 of the
6 respondents decreased as the exercise progressed.
This could have
resulted from a loss of interest or as result of fatigue from the
repetitive nature of the questionnaires and the apparent slow
progress in reaching consensus. In the later rounds of questioning
some experts gave no rationale to support some of their answers or
input. When asked to clarify and support those judgments, they wrote
"See my response in the previous questionnaire," instead of
describing their reasoning again for the group to help improve their
stance.
Coho Decision-Support Model
The resulting coho decision-sport model (Figure 4) is the
outcome of the Delphi process. It presumably represents the best way
of logically organizing the selected parameters by this panel for
evaluating the ecological condition of watersheds relative to their
ability to sustain viable runs of wild coho salmon. The average of
27
Coho Deciston-Suport Model
EMDS Watershed Condition Model for
Oregon Coastal Coho ESU
% Conlfers>5O cm DBH in Watershed
Wats.ihsd UrbAj
%Shadelnrlpsrtai
/
/
1
Vegetation
iiciecc
lj
onIfs 50 an D811 In riparla,
Roads In riparlan
Roads
Stream Crossing CUIVetIS
Road ConnectMty
,-
c.cii cM
zi
Roads on Steep Slopes
Road CondItIon Index
Road Condition
Road Dansity
0;
Oi
Impetvioue Surface
o <'A
WsT
Nutrients
%Bedmd
Substrata <-1k :----4 % Gravil In Riffles
fFInNk Riffles
Physical
Condition
ly
% Slackwater Pools
ecaon ci reZhptr!,t
%Pools
Pools
Deep Pools
\
Reach
Condition
Pieces Large Wood
Wood <i_Y Pieces Large Wood
8iological
rm
Secondary Channel
Condition
hbc
__-
Vetebrata 181
Ec5ccnci
AVE operators pass the average evaluation score
NUN operators pass a score weighted towards the lowest evaluation score
N
nvertetirate 181
ExotIc Fh & Amphtilans
Figure 4. Diagram depicting the coho decision-support model
for the
ENDS system built by experts in a Delphi process to evaluate
the
ecological condition of watersheds for the Oregon Coastal
Coho ESU.
Parameters are enclosed in boxes. Arrows indicate direction
of
aggregation.
28
the panelists' certainty about the coho decision-support model
structure was moderate. Like the AREMP template (Figure 3), watershed
condition is an average of the evaluation scores of the Drivers and
Responses nodes. The Drivers node is an average of the Vegetation and
Roads nodes. The parameters in the Vegetation category include both
riparian and upland indicators. The Roads Condition parameters are
further divided into Road Connectivity and Road Condition categories.
The Response node evaluates to the lowest score of the Water Quality,
Water Quantity, and Reach Condition nodes. Water Quality is divided
into Temperature and Nutrient categories. Like the AREMP model, the
Reach Condition network (orange) feeds directly into the Watershed
Condition knowledge base. The Reach Condition node is an average of
Physical Condition and Biological Condition. Physical Condition is
divided into 4 categories: Channel Incision, Substrate, Pools, and
Wood. All of the terminal nodes in the knowledge base (in boxes) are
parameters where raw data enters the model.
A list of all the parameters in the coho knowledge base, their
definitions, evaluation curves, corresponding evaluation criteria,
data sources, and rationale for use are shown (Table 1)
.
All of the
parameters in the knowledge base (except the road parameters) were
previously used in Oregon's Coho Assessment. They belong to the suite
of field survey and GIS monitoring data collected and managed by
multiple state and federal agencies both before and after the
implementation of the Oregon Plan in 1997. Evaluation curves for each
29
Table 1. Coho decision-support model parameters to evaluate the
ecological condition of watersheds for the Oregon Coastal Coho EStJ.
Parameter definitions, evaluation curves, corresponding evaluation
criteria, data sources and rationale are shown.
Parameter and definition
Evaluation curve
Evaluation Criteria and
source
Data Source and Rationale
Suitable <20%, unsuitable
>40%, (AREMP)
ODF using USGS data
Vegetation
Watershed in Urb/Arg
Percent of watershed area in
urban or agricultural land use.
Metric: %
P*,01oI Wt,th,d In U,tJq
Indicator of sediment input
and runoff potential affecting
flow regime.
%.1n*flflt.flI,aflgt,4nfltInfl
% Shade in Riparian
Percent of 180 degree sky that
is shaded by trees or other
topographic features. Metric:
%
% 91d*
Unsuitable < 76%, suitable
> 91 (ODFW5)
ODFW
Rationale: Influences stream
temperature, and is easily
influenced by human activity.
Conifers > 50cm dbh in Riparian
The number of conifers>
50cm dbh within 30 m of both
sides of the river per 305 m of
primary stream length. Metic:
Confnr '5On, dbh
Unsuitable <22, suitable
>153 (ODFW*)
Rationale: Index of potential
future sources of large wood
recruitment, and highly
responsive to management.
Of canift..
Conifers> 90cm dbh in Riparian
The number of conifers > 90
cm dbh within 30 m of both
sides of the river per 305 m of
primary stream length. Metic:
Conlf
950,,, oh
ODFW
Unsuitable = 0, suitable>
ODFW
79. (ODFW)
Rationale: Index of potential
future sources of large wood
recruitment, highly
responsive to management.
Water Quality
Water Temperature
Evaluation of the 7-day
average maximum water
temperature. Metric: °C
Wat., T.mp.,fln
Phosphorous
Evaluation of total
phosphorous concentration
(mg/L) in the watershed.
PhosphoroLl,
Poor < 7.2°C or >20°C,
good is 10 to 15.5°C. This
crstena is from
California's North Coast
Watershed Assessment
Program (NCWAP) and
was constructed for cohn
and Chinook.
Poor >0.03 mg/L, 25010
percentile of reference
sites. (ODEQ)
Metric: mg/L
CO CCI 010
Nitrogen
Evaluation of the total
inorganic nitrogen
concentration in the
watershed. Metric: mg/L
ODEQ
Rational: Temperature affects
fish distribution, fitness and
physiological processes.
ODEQ
Rationale: essential nutrient
in aquatic ecosystems.
0100 000 CO 031 00100000 III
ToSol no.590010 5101o5,fl
Poor>0.3 mg/L25th
percentile of reference
ODEQ
sites. (ODEQ)
Rationale: essential nutrient
in aquatic ecosystems.
Metric undecided by expert
panel
OWRD
M9IL
Water Quantity
No metric decided by
expert panel
30
Rational: coho salmon need
adequate stream flow to
survive
Physical Condition
Channel Incision
Undecided. Expert panel
did not agree on metric
because data was not found
until after Delphi process
ended.
% Bedrock
Visual estimate of stream
substrate composed of solid
bedrock. Metric: %
% 00d000k
Undecided. Expert panel
did not agree on metric
because data was not found
until after Delphi process
ended.
Poor>ll%, Good <1%.
Rational: Indicator of stream
connectivity to flood plane,
important habitat for juvenile
coho salmon.
ODFW
(ODFW*)
Rational: Measure of lack of
substrate complexity that is
potential factor for decline;
not suitable spawning habitat.
%ar.304.u00nl.
% Gravel in Riffles
Visual estimate of stream
substrate composed of 264mm diameter particle in
riffles. Metric: %
ODFW
Undesirable <26, Desirable
ODFW
>54 (ODFW)
Rational: Measure of
spawning habitat.
%l00010U000M.
% Fines in Riffles
Visual estimate of substrate
composed of <2 mm
diameter in riffles. Metric: %
% POW SOdiment A, RJm.
Unsuitable >22, suitable
<8, (ODFWt)
Rational: Can influence the
survival of eggs and alevins
in the substrate by reducing
oxygenation or physically
preventmg emergence.
% Slackwater Pools
% of pnmary channel area
represented by slackwater
pool habitat (beaver pond,
backwater, alcoves, and
isolated pools). Metric: %
% SIOOP.00000 P0010
Undesirable <0, Desirable
Rational: Measure of channel
morphology, important low
velocity habitat for juveniles.
fc Of oh,00I
Note: Low occurrence of this
variable leads to high
variability in estimate
% P0041
Undesirable <19, Desirable
P00100 lO
Pools> I m deep per km of
primary channel. Metric: #/km
Secondary Channel
% total channel area
represented by secondary
channels. Metric: % area
ODFW
>45 (ODFW)
S
Rational: Measure of channel
morphology, Important low
velocity habitat for juveniles.
0400 font 0040nOd
Deep Pools
ODFW
>7. (ODFW)
area
% Pools
% of primary channel area.
represented by pool habitat.
Metric: % area
ODFW
Undesirable 0, Desirable
ODFW
>4 (ODFW)
Rational: Measure of
channel morphology,
important low velocity
habitat for juveniles.
% 5001d0,ya,.,,01
Undesirable <0.8,
Desirable >5.3 (ODFW')
ODFW
Rational: Measure of channel
complexity, important low
velocity habitat for juveniles.
Note: High vanability m data
31
for this variable results in low
statistical power
Pieces Large Wood
# of pieces of wood> 0.15 m
diameter x 3 m length per 100
m primary stream length.
Metric: # / lOOm
P$... L
Undesirable <8, Desirable
>21 (ODFW*)
Wodi i
Rational: Measure of instream
roughness, and a critical
component of salmon habitat.
Volume Large Wood
Volume of wood>0.15 mx 3
ODFW
LrU. WodVok.n.
Undesirable < 17,
Desirable >58. (ODFW*)
m length per 100 meters
primary stream length.
Metric: m3/ lOOm
ODFW
Rational: Measure of instream
roughness, and a critical
component of salmon habitat.
n,'
mtt.,.
1
Biological Condition
Vertebrate IBI
An index of biotic condition
based on species diversity
that was created for
coldwater streams in
Western Oregon and
Washington. Metric: index
score 1-100.
Macroinvertebrate
River Invertebrate Predication
and Classification System, or
RWPACS. Metric: Uses the
observed over expected (OlE)
score. Index score 0-1.0 EPT.
Exotic Fish & Amphibians
An index of biotic condition
based on species diversity.
Metric: Presence/Absence
Cn,rnnhty $o*
Inve,.te Co
$00..
Exobo F0i, Ond AnWIdb0.,.
H
Poor < 50, 2S percentile
of reference sites (ODEQ
standard)
ODEQ
Rational: Indicator of
biological stream condition.
Poor , 0.9, 2S percentile
of reference sites, (ODEQ
standard)
ODEQ
Absence in fully suitable,
Presence is fully unsuitable
(ODEQ standard)
ODEQ
Rational: Indicator of
biological stream condition.
Rational: Indicator of stream
condition, community
integrity, and competition.
PTtfl,0.t AS0000
* Based on quartiles derived from reference data (optimum OR Plan and Basin reaches within the distribution of coho salmonland use, nparian vegetation, <5% gradient). Source ODFW, unpublished data.
of these parameters were constructed using available information from
ODFW and ODEQ5 where benchmarks had been created to evaluate data for
ODEQ created benchmarks for water quality conditions based on water
quality standards (ODEQ 2005)
For the parameters without established
standards evaluation criteria were made at the 25th percentile of the
distribution of data from reference sites within the ESU (ODEQ 2005)
Similarly, ODFW used quartiles from reference data within the ESU to
establish evaluation criteria for instream and vegetation parameters
(ODFW 2005)
The benefit of constructing evaluation curves from
.
32
the Coho Assessment (Table 1). These benchmarks are evaluation
criteria used to differentiate between suitable and unsuitable
condition and translate directly into evaluation curves. Where
evaluation criteria were either unavailable or insufficient,
alternative sources were used to build evaluation curves.6
The remaining parameters in the coho knowledge base that are shaded
(Figure 4) received strong unanimous support from the expert
panelists to be included in the model because of their important
contribution to the model as a whole, but failed to meet all 3 of the
criteria for inclusion. Their respective reasons for not meeting the
criteria are indicated (Table 2) . These parameters are included in
the decision-support model to indicate data gaps and ideally to
suggest future monitoring needs. No evaluation criteria are given for
the road parameters in Table 1. The parameters Incision and Water
Quantity both met the criteria for inclusion but the panel couldn't
agree on metrics or evaluation criteria. The expert panel agreed that
a Lake parameter should be included in the model but did not agree on
its definition, metric, or placement in the knowledge base.
reference data rather than literature or expert judgment is that they
are specific to the ESU.
6
The Water Temperature evaluation curve was taken from California's
North Coast Watershed Assessment Program (NCWP 2002) because it was
created specifically for coho, whereas ODEQ's temperature benchmark
was a more general standard. Evaluation curve criteria for the
parameter "Percent of Watershed in Urban/Agricultural Land" came from
AREMP because no other exists. From a review of published empirical
studies, the Pew Oceans Commission determined evaluation criteria for
"Percent Impervious Surface" and suggests a formula for determining
the percent of developed land that can be categorized at impermeable
(Beach 2002).
33
Table 2. Parameters that failed at least 1 of the 3 criteria to be
included in the model: (1) They must act as surrogates or indicators
of watershed processes that influence coho; (2) Data for the
parameters much currently exist (in data bases or GIS layers> and be
available to the public; (3) Data for the parameters must exist for
the entire region encompassed by the coastal coho ESU.
Parameter and definition
Cause of eicknion from model
Vegetation
% Large Conifers in Watershed
Data are proprietary (Oregon Forest Industries Council) and not in the public domain
(Dent etal. 2005a).
Roads
Roads in Riparian
No comprehensive GIS data available. BLM has a (MS layer (Ground Transportation
Roads and Trails that could be used but would require significant GIS analysis and
validation.
Stream Crossing Culverts
No comprehensive GIS data available. BLM has a GIS layer (Ground Transportation
Roads and Trails that could be used but would require significant (MS analysis and
validation.
Roads on Steep Slopes
No comprehensive GIS data available. BLM has a (ITS layer (Ground Transportation
Roads and Trails that could be used but would require significant (ITS analysis and
validation.
Road Condition Index
Oregon Department Forestry only has data available for 2 watersheds.
Road Density
No comprehensive GIS data available. BLM has a GIS layer (Ground Transportation
Roads and Trails that could be used but would require significant GIS analysis and
validation.
Impervious Surface
No comprehensive GIS data available. Significant GIS analysis and validation required.
After 4 rounds of questionnaires the following parameters were
still being debated as to whether or not they should be included in
the model: Percent Public Land in Watershed, 303d Listing, High
Intrinsic Potential (HIP), Geology, Gradient, Morphology, and Bedrock
Substrate.
The aggregation functions (Appendix 1) that determine how nodes
are combined and passed on to subsequent nodes are shown at each
junction in arrow shaped boxes (Figure 4). Unlike the AREMP template
34
that used only AVE operators the expert panelists chose to use MIN
operators for aggregating scores to following categories: Vegetation,
Roads, Road Connectivity, Road Condition, Water Quality, Responses,
Physical Condition and Wood. The MIN operator passes on the minimum
scores from those nodes being aggregated.
Although consensus was
achieved for operator choice throughout the model, only 1 of the
expert panelists expressed total confidence in their choices.
Individual panelists expressed moderate to high uncertainty regarding
the selection of their own aggregation functions.
The expert panel chose not to weight any of the parameters in
the model. No one felt comfortable making explicit recommendations
without being able to manipulate the complete computer model because
of uncertainty about how the model would respond. Instead the panel
agreed on 3 parameters that should be weighted in the final model
based on their expert judgment and relevant literature (Reeves et al.
1989, Nickelson et al. l992a, Mickelson et al. 1992b, Sharma and
Hilborn 2001, and Pess et al. 2002)
.
These are:
(1) Temperature;
(2)
Pools (specifically Secondary Channels and Slackwater Pools); and (3)
the Wood category.
DISCUSSION
Delphi Process
Although the Delphi technique helped achieve the goal of
creating the coho decision-support model, whether it added
35
appreciable value to the outcome requires analyzing some of the
difficulties of the process and concerns of its use.
The Delphi technique was selected for this case study primarily
because bringing experts together from all over the Pacific Northwest
would have been costly and logistically challenging. Administering
the Delphi process via mail allowed experts to participate from
distant locations from southern Oregon to northern Washington. This
logistical benefit was also one of the Delphi technique's greatest
drawbacks in that it slowed the process. It took 4 questionnaires and
over 8 months to create the coho decision-support model. Crance
(1987) found that generally 4 questionnaires are submitted before the
process is complete. AREMP's preliminary ENDS knowledge base was
built by experts over a 3-day group work session (Gallo, personal
communication).
The largest obstacle in establishing parameter weights and
choosing aggregation functions was that experts didn't have access to
a functional computer model. This limited experts' abilities to make
recommendations and offer suggestions. A group work session with
computers, ENDS software, a draft model, and an ENDS expert could
have reduced these difficulties and the resulting uncertainty but
would have greatly increased the cost.
The Delphi technique is not well suited to building consensus
on such complex topics as knowledge-based decision-support model
structure. Highly complex subjects and open-ended questions can lead
36
to problems with transferring and quantifying feedback (Rowe and
Wright 1999). After reviewing a cross section of published Delphi
studies, Rowe and Wright (1999) determined that successful Delphi
processes consist of questions that elicit numerical estimates about
a topic. Quantitative feedback (e.g., averages, estimates, and
probabilities) allows statistical aggregation of responses. Rowe and
Wright (199) concluded that more study is needed to define
appropriate topics and questions for the Delphi technique.
Creating the coho decision-support model structure required
open-ended questioning. The model network contains a large quantity
of information including parameters, rules for combining them, inter-
parameter relationships, and weights. Dozens of ways are possible to
organize the large number of parameters. While each expert may have a
similar understanding of watershed condition indicators and
ecological processes, each conceives watershed condition in their own
unique mental picture. This mental picture is difficult to clarify
and articulate. Suggesting dramatic changes to the model structure
was challenging to rationalize and communicate to other expert
panelists. When changes in model structure were suggested that
differed substantially from the AREMP model, they were difficult for
the moderator to understand (e.g., due to little explanation or
rationale) and a challenge to add to the subsequent questionnaire in
a way that would elicit a useful response. In each case where these
changes were proffered they were unanimously rejected. This may have
indicated poor communication in the Delphi process or uncertainty
with straying from the established AREMP model structure.
Making
37
large changes in the model structure would have required more
questionnaires in the Delphi process because of the slow transfer of
information relative to a group setting. Large changes in model
structure could have been made more effectively in a group setting
where the model structure could be drawn and edited more quickly.
Dai et al.
(2004) reminds us of the fact that it is the experts
that ultimately create the model in any expert-based modeling
exercise. Therefore in theory, a different expert panel with a
different number of panelists would have created a different model.
Rowe and Wright (1996) found that more accurate experts changed their
answers less over rounds in the Delphi process than those who were
initially less accurate or less "expert". This may affect the outcome
of a Delphi process and highlights the importance of the expert
selection process.
The moderator has a certain amount of leverage, control and
influence in the Delphi Process. Most of the literature focuses on
the objective role of the expert panelists but the potential of the
moderator to influence the process is seldom addressed.
For example,
when summarizing information from responses to present in the
following questionnaire, the moderator had to be selective about what
information would be most relevant and useful and what information
would be distracting. Sorting through these comments cost more
administrative time. Clayton (1997) points out that one of the values
of the Delphi technique is being able to eliminate irrelevant
information from the feedback process before it spawns time-consuming
38
tangents that are common in group discussions. This ability to
provide controlled feedback has the potential to be abused. A biased
moderator or even one under a time deadline could direct the outcome
either intentionally or unintentionally. Therefore credibility may be
just as important for the moderator as it is for the expert
panelists.
One common problem in group decision-making that the Delphi
technique attempts to remedy is the undue influence of dominant
individuals. Dominant participants through their language and social
behavior can control and direct the outcome of group decision making
exercises in face-to-face settings. In some instances, experts that
expressed dominant and confident ideas initially deferred to others
later in the process. This occurred in later rounds of the exercise
after multiple attempts to weight certain model parameters. After two
rounds of apparently confident input one panelist wrote: "I defer to
the others who might have more expertise in this specific area." As a
consequence, those experts whose opinions endured, won out in the
end.
The overall quality of the responses in a Delphi exercise is
influenced by the interest and commitment of the expert panelists
(Delbecq et al. 1975, Crance 1987, and Clayton 1997) .
Making the
questionnaire a priority over other professional and personal
obligations requires high motivation since other people are not
present (Crance 1987) . This may reduce a panel member's efforts to
consider thoroughly and respond completely to all the elements of the
39
questjonnajr
and influence the final outcome (Clayton 1997)
.
It is
important to note that all panelists
were volunteers. Had they been
paid or required by their employers to
participate, the detail and
timeliness of their responses may have differed.
Imprecise communication impeded the consensus
building process
and frustrated both the moderator and
panelists. One example that was
debated over 3 questionnair
was the differences between watershed
potential, watershed suitability, and watershed
condition. Some
participants saw these as vastly separate entities
and were reluctant
to continue in the Delphi exercise until they
were resolved and made
explicit. Other participants expressed no concerns and
were
apparently unaffected by the discussion. A potential remedy could
have been a training session implemented before the
Delphi process to
cover not only definitions but model background and function
as well.
This could have been accomplished by the moderator traveling
to the
panelists individually or via phone.
The final coho decision-support model is similar in structure
to the IAREMP model. Both models have the same initial
categories and
similar arrangement. Providing the AREMP model
as a template in the
Delphi process likely influenced this outcome. The AREMP model
was
originally provided to illustrate to the panelists an example of
a
proven decision-support model. In future studies we recommend
providing additional model Options as learning tools to decrease the
focus on this individual model.
40
Building unbiased consensus was the ultimate goal of the Delphi
process, but should consensus really be the goal? Keith (1996) argues
that although valuable for forming a credible best estimate of
current scientific knowledge, combining experts' judgments is
undesirable. Consensus gives the illusion that all members in the
decision making process are satisfied, masking any conflicts or
differing opinions. To create the coho knowledge base all members
presented unique and sometimes stubborn judgments but were forced to
compromise during the process. A majority approval of an alteration
or addition to the coho decision-support model resulted in a
permanent change even when one or two individuals were in
disagreement. Peterson et al.
(2005) describes consensus as fatal to
democratic decision making because differences and conflicts are
obscured and final consensus is perceived as the ultimate truth.
Keith (1996) warns that certain methods of building consensus produce
answers that are acceptable to all rather than seeking the most
correct answers. Whether or not the coho knowledge base is viewed as
the best answer considering the best current science and expert
judgment, it must certainly be viewed as just one alternative that
can and will change as new judgments and information are added.
Utility of EMDS
The effective use of ENDS to evaluate watershed condition for
coho in Oregon's Coastal Coho ESU was severely constrained by lack of
data (Table 3) . For some parameters no data are available, for others
the data are privately owned, and for many of the Driver parameters
41
the data are GIS based and require analysis. The majority of the data
that are available are statistically robust enough to make
assessments at the ESU or monitoring area levels but not at the
population (basin) level. Data consistency is also an issue. Multiple
state and federal agencies, private companies, and advocacy groups
are involved in monitoring efforts for the Coastal Coho ESU. They all
collect different data at different spatial extents and resolutions,
and using different methods, varying year to year. Since the Coho
Assessment, OWEB created and maintains a data warehouse to make data
more accessible, but determining data availability, coverage,
sources, and who to contact for more specific information is still
difficult.
Table 3. Insufficient data exists to make watershed assessments at
the population scale. The following shows which categories of
parameters have suitable data for assessments at each of the three
scales: ESU, Monitoring area, and population.
Parameter Categories
Data Source
Vegetation parameters
Water Quality
Scale
Population
ESU
Monitoring Unit
ODFW'
Yes
Yes
ODEQ2
Yes
Yes
Water Quantity
OWRD3
Yes
Yes
Yes
Physical Condition
ODFW
Yes
Yes
Select basins
Biological Condition
ODEQ
Yes
Yes
Roads
BLM4
-
'Oregon Department of Fish and Wildlife
2Oregon Department of Environmental Quality
3Oregon Water Resources Department
4Bureau of Land Management (potential source of GIS data)
Both the panel of experts in this case study and the authors of
the Oregon coastal coho assessment rated roads as one of the primary
drivers of watershed condition that influence coho.
One of the goals
42
of the Oregon Plan is to reduce the effects of roads on streams and
aquatic habitat (Mills et al. 2005) . As outlined by Mills and others
(2005), the primary effects of roads on streams and coho include
restricting fish passage, delivering sediment, altering aquatic
habitat, and changing hydrology and stream flow.
Several road parameters for the model were suggested by ODF
experts (Mills and Dent per communication) that were consistent with
recommendations of the Independent Multidisciplinary Science Team
(IMST)
.
These were the density of stream crossing culverts, road
density in riparian areas, and road density on steep slopes.
Comprehensive, high-quality data for these road parameters are
unavailable at this time. The BLM has a GIS layer (Ground
Transportation Roads and Trails) that could be used to find road
density in both riparian and steep hill slope areas, but this Would
require significant GIS analysis and, the road data are inconsistent
and lacking among different land ownerships. To make the layer
useful, a ground survey would be necessary to validate missing roads
and calibrate for the quality of roads (Mills per verbal
communication). As part of the Oregon Plan a program to monitor roads
consistently across forest ownerships has been proposed but not yet
implemented.
One parameter that was suggested by the coho decision-support
model expert panel members and used in the past by others as an
indicator of the effects of roads on streams is overall watershed
road density. But Mills and others
(2005) caution that this is not a
43
reliable indicator unless all roads are in similar condition,
location, and built with the same practices.
A potential future parameter for the model might be road
condition. Oregon Department of Forestry has developed a Road
Information Management System with an explicit procedure for
evaluating road condition based on numerous factors including the
relative risk of influencing aquatic and riparian habitat. To date
this has only been used for two watershed analyses on state forest
roads but would be more useful if expanded to all land ownerships and
watersheds across the coho ESU.
One prominent factor that inhibits the wide spread use and
acceptance of ENDS is skepticism among experts. Use of the ENDS
system for assessment of ecological condition is relatively new and
unexplored. AREMP is the only organization that has used and
continues to use ENDS successfully to assess watershed condition
(Gallo et al. 2005)
.
The NCWAP experts who built, and are still
refining their initial model, voiced skepticism about the model
function, abilities, usefulness, and rigor (NCWAF 2002)
.
One expert
on their team called it a "yet-to-be validated working hypothesis of
the factors that define watershed condition" (NCWAP 2002)
.
The
RBRTCT's assessment of ENDS to identify and prioritize restoration
needs in the Applegate Sub-basin lends some credence to the concerns
of the NCWAF team. The RBRTCT concluded that using ENDS to average
scores from causal processes and instream response variables did not
improve their ability to accurately evaluate watershed condition or
44
provide management direction (RBRTCT 2004)
.
They found that their
model was more useful when broken down into smaller clusters of
parameters.
The experts working on the coho decision-support model also
expressed skepticism throughout the Delphi process. This revolved
around model structure, parameter relationships, and the ability of
the knowledge base to provide more than a theoretical schematic of
watershed condition. The following are typical comments from the
Delphi questionnaires.
It seems that the drivers, which should influence the response
variables are disconnected from the response variables in the
model. What happens if the response variables don't follow the
drivers?
The structure of the model is seriously flawed. The model lacks
integration of the natural capacity of a watershed to produce
coho salmon.
I do find the overall structure of the model a little odd as I
usually think of drivers of habitat broad-scale features such
as geology, morphology, etc., with more specific factors
influences reach level productivity...Your model seems to have
repeated the drivers of basic potential and habitat condition
which I don't really understand though perhaps I'm not clear on
how the model works.
Although the coho decision-support model provided these experts
with an opportunity to explicitly conceptualize watershed condition
relative to coho, demonstrating the utility of EMDS will require
validation, repetition, and continual peer review.
The Oregon Coho Assessment states that .T.REMP's use of EMDS to
characterize the ecological condition of watersheds and aquatic
45
ecosystems for the area managed under the Northwest Forest Plan is
"broadly applicable to federal forestland in the Coastal Coho ESU"
(Nicholas et al. 2005)
.
Given this and the list of strengths above,
why did not Oregon use an ENDS watershed assessment model for at
least some part of the 2005 Coho Assessment? When asked ODFW offered
four reasons:
(1) ODFW in not a land-management agency;
(2)
Generally, the stream information that ODFW collects and uses is at a
smaller reach scale;
(3) ODFW's assessment methods are more fish
centric opposed to habitat centric; and (4) there are no available
funds to implement a new model that would require software, skilled
personnel, and potentially a different and more rigorous sampling
effort.
Additional work is needed to make the current version of the
coho decision-support model a useful and functional tool for
watershed assessment. The model must be exposed to a broader group of
experts for their edits and recommendations. The model could then be
implemented into the ENDS system and run with real, or in the case of
the Road parameters, mock data. The influence of each of the
parameters to the overall output could be assessed. Parameters that
have little influence or duplicate another parameter's influence
could potentially be removed. If the model performed satisfactorily,
and provided useful information about watershed condition, a decision
would need to be made about whether to use part(s) of model alone,
use the model without the road parameters, or invest in acquiring the
necessary road parameters for a comprehensive model. If long-term use
46
of the model was desired, a monitoring
effort guided by the specific
needs of the coho decision-support model
would be ideal.
CONCLUSION
At the population level the EMDS system
was not useful for
assessing watershed condition for coho due to
the lack of data. As a
whole, data for a majority of the
parameters in the coho decisionSupport model were robust enough for assessments
at the extent of the
ESU and at the monitoring area. In addition,
data for all of the road
parameters in the knowledge base were either
nonexistent, or not
available because they are privately owned or require
GIS analysis.
These findings prevented completing and running
the coho knowledge
base in the EMDS system. If the data did exist,
or if it is gathered
during future monitoring efforts, EMDS might be
valuable to those
managing the Esu, especially since the fundamental coho
knowledge
base exists as well as the evaluation curves for the
majority of
parameters based on empirical data (reference conditions)
and
literature.
EMDS is an objective, transparent, and repeatable
watershed
assessment tool (Bleier et al. 2003, Dai et al. 2004, Reeves et
al.
2004, Gallo et al. 2005, Reynolds and Hessburg 2005.)
But some
experts, including members of the panel on this
study, are skeptical
of the abilities of EMDS. As AREMP's program matures,
the usefulness
of EMDS may become more credible and acceptable for
use in other
wicked problems like salmon recovery. Even now it has potential
47
because, at the very least, the coho decision-support model helps
experts to define watershed condition for coho, points out data gaps,
monitoring needs, and provides a framework for generating new
hypothesis and testing assumptions about the relationships among
watershed parameters.
Based on the case study reported here, eliciting expert
judgment through a Delphi process does not add appreciable value to
creating a decision-support model for watershed assessment. The
Delphi technique is not a foolproof method of eliminating bias or
personal values and should not be assumed to be inherently superior
to more informal methods. Normative science can still enter into the
system from the moderator or from dominant individualsparticularly
when building consensus on uncertain topics where some experts
default to others when they perceive the apparent confidence of
others as more certain than their own knowledge.
In addition the coho decision-support model is complex and not
suitable to being created through a Delphi process due to the slow
nature of the process via mail and overall inefficient transfer of
information in a questionnaire/response format. The results could
have been generated in a single day-long group session using a
Nominal Group technique. All of the time saved could have been more
valuable than preserving the anonymity of the panelists plus the cost
of bringing everyone together.
48
Assessments can be no better than the information on which they
are based (Lawrence et al. 1997)
.
Although in theory different
experts would have each created a different model, we believe that
the dominant driver for how the coho decision-support model turned
out was probably the AREMP template that was provided to the experts
in the beginning. The similarity between the coho knowledge base and
the AREMP knowledge base seemed predictable. In the end, some of the
most useful insights from the experts were the areas of high
uncertainty or no consensus. These areas, like High Intrinsic
Potential (HIP), suggest important and necessary areas of focused
future research because experts were either unfamiliar with them or
weren't convinced that they would be appropriate for use in the coho
decision-support model.
We conclude that in the case of the Oregon Coastal Coho EStJ,
using formal methods of eliciting and applying expert judgment have
many limitations and added marginal value. Compared with group
meetings the Delphi technique is relatively inefficient and
impractical for providing information to decision makers and dealing
with complex, urgent, wicked problems like salmon recovery. The EMDS
system is currently unusable at the preferred spatial scale (the ESU)
but provides a way to build consensus and understanding about
watersheds, their processes, and influences on coho salmon. If the
misssing data were available the coho decision-support model would
function in the EMDS system but would still need to pass muster among
a more extensive group of experts and decision-makers before being
adopted by any agency for use.
49
Given the plethora of wicked natural resource issues that
exist, there will be a continued reliance
on experts to provide
crucial information to decision makers. To be most useful expert
judgment must be relevant, credible, timely, and
communicated
effectively. The formal methods of eliciting and applying
expert
judgment should aid in meeting these criteria.
In this case study we
analyzed the tradeoffs between these formal methods in that
improving
credibility and transparency came at the cost of time and
procedural
efficiency. FoLmal methods of eliciting and applying
expert opinion
for assessments are no panacea. Managers and decision
makers will
need to weigh the pros and cons on a case by
case basis when
contemplating whether these tools will add appreciable
value to their
decision-making.
50
BIBLIOGRAPHY
Beach, D. 2002. Coastal sprawl: the effects of urban design on
aquatic ecosystems in the United States. Pew Oceans Commission,
Arlington, Virginia.
Bisbal, G. A. 2002. The best available science of the management of
anadromous salmonids in the Columbia River Basin. Canadian Journal
of Fisheries and Aquatic Sciences 59: 1952-1959.
Bisbal, G. A. 2006. Learning to decide and deciding to learn:
conduits to wild salmon in 2100? Pages xx-xx in R. T. Lackey, D.
H. Lach, and S. L. Duncan, editors. Salmon 2100: the future of
wild Pacific salmon. American Fisheries Society. Bethesda,
Maryland.
Bleier, C., S. Downie, S. Canriata, R. Henly, R. Walker, C. Keithley,
M. Scruggs, K. Custis, J. Clements, and R. Klamt. 2003. North
Coast Watershed Assessment Program (NCWAP) Methods Manual.
California Resources Agency and California Environmental
Protection Agency, Sacramento, California.
http: / /www. ncwap . ca . gov/pdf /NCWAP Methods Manual Final . pdf
Bogert, L. M. 1994. That's my story and I'm stickin' to it: is the
"best available" science any available science under the
Endangered Species Act? Idaho Law Review 31: 85-150.
Clayton, H. J. 1997. Delphi: a technique to harness expert opinion
for critical decision-making tasks in education. Educational
Psychology: An International Journal of Experimental Educational
Psychology 17(4): 373-386.
Cleaves, D. A. 1994. Assessing uncertainty in expert judgments about
natural resources. General Technical Report SO-1 10. U.S.D.A.
Forest Service Southern Forest Experiment Station, New Orleans,
LA. July 1994.
Crance, J. H. 1987. Guidelines for using the Delphi technique to
develop habitat suitability index curves. Biological Report
82(10.134). National Ecology Center, U.S. Fish and Wildlife
Service, Fort Collins, Colorado.
Dai, J. J., S. Lorenzato, and D. N. Rocke. 2004. A knowledge-based
model of watershed assessment for sediment. Environmental Modeling
& Software 19: 423-433.
Delbecq, A. L., A. H. Van DeVen, and D. H. Gustafson. 1975. Group
techniques for program planning-a guide to nominal group and
Delphi processes. Scott Foresman and Company, Glenview, IL.
Dent, L., and A. Herstrom. 2005a. Land use and land cover
characteristics in the Coastal Coho ESU. Oregon Plan Assessment
51
Part 4J OP Technical Report 3.
http: //nrimp. dfw. state. or. us/OregonPlan/
Ellison, A. M. 1996. Pn introduction to Bayesian inference for
ecological research and environmental decision-making. Ecological
Applications 6(4): 1036-1046.
Gallo, K., S. H. Lanigan, P. Eldred, S. N. Gordon, and C. Moyer.
2005. Northwest Forest Planthe first 10 years (1994-2003):
preliminary assessment of the condition of watersheds. General
Technical Report PNW-GTR-647. Portland, OR: U.S. Department of
Agriculture, Forest Service, Pacific Northwest Research Station.
Gordon, S., and K. Gallo. 2002. EMDS decision support model approach
the evaluation of watershed condition. Prepared for the StateFederal Monitoring Program. Northwest Forest Plan: Regional
Ecosystem Office Monitoring Publications Page.
http: //www.reo.gov/monjtorjng/watershed/docs/Al2O6O29sGordonDsM
white paper .pdf.
Governors Natural Resource Office (GNRO). 2005. Oregon coastal coho
assessment, part 3a: policy for the evaluation of conservation
efforts; the certainty that the conservation effort will be
implemented. Salem, Oregon.
Independent Science Advisory Board (ISB). 2003. Decision support
models as tools for developing management strategies: examples
from the Columbia River Basin, Chapter 10. Pages 233-242 in R. C.
Wjssmar and P. A. Bisson, editors. Strategies for Restoring River
Ecosystems: Sources of Variability and Uncertainty in Natural and
Managed Systems. \merican Fisheries Society, Bethesda, Maryland.
Jensen, M. E., K. Reynolds, J. P.ndreasen, and I. Goodman. 2000. A
knowledge-based approach to the assessment of watershed condition.
Environmental Monitoring and Assessment 64: 271-283.
Johnson, C. J. and M. Gillingham. 2004. Mapping uncertainty:
sensitivity of wildlife habitat ratings to expert opinion. Journal
of Applied Ecology 41: 1032-1041.
Keith, D. W. 1996. When is it appropriate to combine expert
judgments? Climate Change 33: 139-144
Lackey, R. T. 2004. Normative science. Fisheries 29(7): 38-39.
Lawrence, F. A., J. J. Stone, P. Bellman, and L. J. Lane. 1997. Using
measured data and expert opinion in a multiple objective decision
support system for semiarid rangelands. Transactions of the ASAE
40(6): 1589-1597.
Lee, D. 2000. Assessing land-use impacts on bull trout using Bayesian
belief networks. Pages 127-147 in S. Ferson and M. Burgman,
editors. Quantitative Methods for Conservation Biology. Springer,
New York.
52
Magnuson, J. J., F. W. Allendorf, R. L. Beschta, P. A. Bisson, H. L.
Carson, D. W. Chapman, S. S. Hanna, A. R. Kapuscinski, K. N. Lee,
D. P. Lettenmaier, B. J. McCay, G. N. McNabb, T. P. Quinn, B. E.
Ridell, and E. E. Werner (Committee on Protection and Management
of Pacific Northwest Anadromous Salmonids, National Research
Council) . 1995. Upstream: Salmon and Society in the Pacific
Northwest, National Academy Press, Washington.
Meengs, C. C., and H. T. Lackey. 2005. Estimating the size of
historical Oregon salmon runs. Reviews in Fisheries Science 13:
51-66.
Mills, K., L. Dent, J. Paul, and B. Riggers. 2005. Reducing effects
of roads on salmonids under the Oregon Plan. Oregon Plan
Assessment Part 4J, Technical Report 1.
http: //nrimp . dfw. state .or. us/OregonPlan/
Nickelson, T. E., J. D. Rodgers, S. L. Johnson, and M. F. Solazzi.
1992a. Seasonal changes in habitat use by juvenile coho salmon
(Oncorhynchus kisutch) in Oregon coastal streams. Canadian Journal
of Fisheries and Aquatic Sciences 49(4): 790-794.
Nickelson, T. E., N. F. Solazzi, S. L. Johnson, and J. D. Rodgers.
1992b. Effectiveness of selected stream improvement techniques to
create suitable summer and winter rearing habitat for juvenile
coho salmon (Oncorhynchus kisutch ) in Oregon coastal streams.
Canadian Journal of Fisheries and Aquatic Sciences 49(4): 790-794.
Nicholas, J., B. Macintosh, and E. Bowles. 2005. oregon coastal coho
assessment, part 1: synthesis of the Oregon coastal coho
assessment. Oregon Watershed Enhancement Board and Oregon
Department of Fish and Wildlife, Salem, Oregon.
ftp://nrimp.dfw.state.or.us/oregonplan/reports/FinalReports/
North Coast Watershed Assessment Program (NCWAP). 2002. EMDS Science
and Practitioner Review Panel. Unpublished data.
http: //www. ncwap. ca. gov/pdf/EMDS Review Report .pdf
Oregon Department of Environmental Quality (ODEQ) 2005. Oregon plan
for salmon and watersheds, Oregon coastal coho assessment water
quality report, Part 4(B) DEQ Water Quality. Oregon Department of
Environmental Quality, Salem, Oregon.
http: //nrimp .dfw. state .or.us/OregonPlan/
Oregon Department of Fish and Wildlife (ODFW). 2005. Factors for
decline and break points (unpublished data). Oregon Department of
Fish and Wildlife, Salem, Oregon.
Oregon Watershed Enhancement Board (OWEB). 2005. 2003-2005 Oregon
plan biennial report, volume 2. Oregon Watershed Enhancement
Board, Salem, Oregon.
53
Pess C. R., D. R. Montgomery, R. B. Bilby, E. A. Steel, R. E Bilby,
B. B. Feist, and H. N. Greenberg. 2002. Influence of landscape
characteristics and land use on coho salmon (Oncorhynchus kisutch)
abundance in the Snohomish River, Washington State, USA. Canadian
Journal of Fisheries and Aquatic Sciences 59: 613-623.
Pess, G. R., T. J. Beechie, J. E. Williams, D. R. Whitall, J. I.
Lange, and J. R. Klochak. 2003. Watershed assessment techniques
and the success of aquatic restoration activities. Pages 185-201
in R. C. Wissmar and P. A. Bisson, editors. Strategies for
Restoring River Ecosystems: Sources of Variability and Uncertainty
in Natural and Managed Systems. .merican Fisheries Society,
Bethesda, Maryland.
Peterson, N. N., M. J. Peterson, and T. R. Peterson. 2005.
Conservation and the myth of consensus. Conservation Biology
19(3): 762-767.
Rauscher, N. H. 1999. Ecosystem management decision support for
federal forests in the United States: a review. Forest Ecology and
Management. 114: 173-197.
Regan, T. J., L. L. Master, and G. A. Hammerson. 2004. Capturing
expert knowledge for threatened species assessments: a case study
using MatureServe conservation status ranks. Acta Oecologica 26:
95-107.
Reeves, G. H., D. B. Hohler, D. P. Larsen, D. B. Busch, K. Kratz, K.
Reynolds, K. F. Stein, T. Atzet, P. Hays, and N. Tehan. 2004.
Effectiveness monitoring for the aquatic and riparian component of
the Northwest Forest Plan: conceptual framework and options.
General Technical Report PNW-GTR-577. U.S. Department of
Agriculture, Forest Service, Pacific Northwest Research Station.
Portland, Oregon.
Reeves, G. H., F. H. Everest, T. B. Nickelson. 1989. Identification
of physical habitats limiting the production of coho salmon in
western Oregon and Washington. General Technical Report USDA
Forest Service Pacific Northwest Range Experimental Station.
Reynolds, K. N., and G. H. Reeves. 2003. Role of knowledge-based
systems in analysis and communication of monitoring data. Pages
189-210 in D. B. Busch, and J. C. Trexler, editors. Monitoring
Ecosystems: Interdisciplinary Approaches for Evaluating
Ecoregional Initiatives. Island Press, Washington.
Reynolds, K. N., and G. H. Reeves. 2004 Analysis and communication of
monitoring data with knowledge-based systems. Pages 95-104 in H.
Gucinski, C. Miner and B. Bittner, editors. Proceedings: Views
From the Ridge-Considerations for Planning at the Landscape Scale,
Vancouver, WA. General Technical Report. PNW-GTR-596. ILS.
Department of Agriculture, Forest Service, Pacific Northwest
Research Station. Portland, Oregon.
54
Reynolds, K. M., and P. F. Hessburg. 2005. Decision support for
integrated landscape evaluation and restoration planning. Forest
Ecology and Management 207: 263-278.
Reynolds, K. M., N. Jensen, J. Andreasen, and I. Goodman. 2000.
Knowledge-based assessment of watershed condition. Computers and
Electronics in Agriculture 27: 315-333.
Reynolds, K. M., S. Rodriguez, and K. Bevans. 2002. Ecosystem
management decision support 3.0: user guide. USDA Forest Service,
Pacific Northwest Research Station, Corvallis, Oregon.
Rogue Basin Restoration Technical Core Team (RBRTCT)
2004.
Ecological management decision support analysis of the Applegate
sub-basin, final report.
.
http: //www. restoretherogue . org/model/index. html
Rowe, G., and G. Wright. 1996. The impact of task characteristics on
the performance of structured group forecasting techniques.
International Journal of Forecasting 12: 73-89.
Rowe, G., and G. Wright. 1999. The Delphi technique as a forecasting
tool: issues and analysis. International Journal of Forecasting
15: 353-375.
Rykiel, E. J. 2001. Scientific objectivity, value systems, and
policymaking. Bioscience 51(6): 433-436.
Schmoldt, D. L., J. Kangas, and G. Mendoza. 2001. Basic principles of
decision making in natural resources and the environment. Pages 1133 in D. L. Schmoldt, J. Kangas, G. A. Mendoza, and H. Pesonen,
editors. The Analytic Hierarchy Process in Natural Resource and
Environmental Decision Making. Kiuwer Academic Publishers, Boston.
Sharma, R., and R. Hilborn. 2001. Empirical relationships between
watershed characteristics and coho salmon (Oncorhynchus kisutch)
smolt abundance in 14 western Washington streams. Canadian Journal
of Fisheries and Aquatic Sciences 58: 1453-1463.
Stanford, J. A., and 3. V. Ward. 1992. Management of aquatic
resources in large catchments: recognizing interactions between
ecosystem connectivity and environmental disturbance. Pages 91-124
in R. J. Naiman, editor. Watershed Management: Balancing
Sustainability and Environmental Change. Springer-Verlag, New
York.
Tversky, A., and D. Kahneman. 1982. Judgment under uncertainty:
heuristics and biases. Pages 3-22 in D. Kahneman, P. Slovic, and
A. Tversky, eds. Judgment Under Uncertainty. Cambridge University
Press, Cambridge.
55
APPENDICES
56
APPENDIX 1.
Ecosystem Management Decision Support (EMDS)
In the knowledge-based decision support modeling approach,
watershed assessment is a multi-criteria evaluation in which experts
select the parameters that characterize the watershed, define their
evaluation criteria, and determine the relations between those
parameters. A simplified diagram of the structure and function of the
knowledge base is shown in Appendix 1 Figure 1. It resembles a tree
diagram similar to a flow chart where parameters that act as
surrogates or indicators of watershed processes that influence
overall watershed condition are arranged hierarchically based on
current scientific understanding of their processes, and logical
relationships with one another.
Parameters are represented by nodes
where connections and direction of influence are indicated by arrows.
The complete knowledge base can be thought of a mental map of logical
dependencies among watershed parameters (Reynolds and Hessburg 2005).
The terminal nodes on the right are the individual watershed
parameters (e.g. stream temperature, pool frequency, riparian canopy,
etc.) where raw data enter the model for evaluation against pre-
established reference conditions in fuzzy logic curves (herein
evaluation curves) (Appendix Figure 2). Fuzzy logic is a formal branch
of mathematics that deals with quantifying imprecise parameters and
their relationships with other parameters. Its use is ideal in
ecosystem evaluation because ecosystems and their components have few
specified points where "good" condition turns to "poor" condition
57
Evauate/
+ Aggregate
Combined Score
Normahze
Watershed
Parameters (data)
Watershed
Condition
Score
4
4
Watershed
Appendix 1 Figure 1. Conceptual diagram of an EMDS knowledge base to
evaluate watershed condition.
(Reeves et al. 2004)
.
Instead conditions generally transition along a
gradient which is what fuzzy logical tries to display. Fuzzy curves
are similar to habitat suitability curves in that they define the
benchmarks or break points for good and poor condition.
Each evaluation curve defines the criteria that the data must
meet in order to fully support the proposition of the model: that the
attribute is in good condition (or in this case, that the condition
of the attribute is suitable for sustaining viable populations of
coho) .
Evaluation curves are constructed based on a combination of
available literature, empirical evidence, and expert judgment. Data
58
is then evaluated and normalized based on this criterion and given a
score between +1 to -1. A score of +1 indicates that the parameter is
in good condition or is fully suitable to sustain viable runs of wild
coho. Conversely, a score of -1 indicates the parameter is fully
unsuitable. Evaluation scores between +1 and -1 reflect the gradient
between good and poor conditions, Converting scores to this simple,
common scale of +1 to -1 facilitates comparing and aggregating scores
later in the model (Gordon and Gallo 2002)
Water Temperature
6
8
10
12
14
16
18
20
22
24
26
oc
Appendix 1 Figure 2. This evaluation curve represents the evaluation
criteria for water temperature. This curve is based on literature and
empirical data for coho and Chinook salmon and was created by experts
from California's North Coast Watershed Assessment Program (NCWAP)
Here if water temperature is between 10.0 and 15.5°C it is evaluated
as fully suitable to sustain coho (+ 1.0) or in "good condition."
Temperatures cooler or warmer than this window would receive
declining scores.
After evaluation, individual model parameter scores are
aggregated hierarchically at multiple junctions. For example, an
intermediate level node may combine multiple parameters into general
categories of instream, riparian, and upland conditions.
Any given
node assesses the proposition that all the factors leading to it are
in suitable condition at that level. Ultimately all parameters scores
in the network are aggregated logically into a single score of
59
watershed condition (between +1 and -1), reflecting its condition
relative to coho suitability.
At each junction in the network where two or more parameters
are aggregated, expert defined rules determine how scores are
aggregated and passed on to the subsequent level.
These aggregation
functions, called logic operators, can pass on the minimum score,
average score, or maximum score from those being combined. Minimum
"MIN" operators pass a score weighted towards the lowest evaluation
score. MIN is used primarily to allow one indicator to override other
indicators. Average "AyE" operators pass on the average evaluation
score. AVE is used so that indicators in good and poor condition
balance each other out. Maximum "MAX" passes the highest evaluation
score and therefore presents an optimistic view of condition. These
operators, like the configuration of the model, are based on current
scientific understanding of the parameter, their processes, and
logical relationships with one another. In addition to operators each
node in the model can be assigned a weight.
Model output comes in the form of tables, graphs, and color
coded maps (Appendix Figure 3) displaying watershed condition scores
or parameter influence rank to the overall model. Scores (between +1
and -1) are generally are broken into a 7 classes for ease of
comparison and to improve understanding and communication of results
(Appendix Table 1).
60
'satershed condition
-I (False)
-Ito -.5
-.5 to 0
0(No Data)
o to .5
.5 to I
I (True)
Appendix 1 Figure 3. An example map of ENDS model evaluation results
showing color coded rankings of watershed condition. Source: EMDS
website, http://www.fsl.orst.edu/emds/.
Appendix 1 Table 1. EMDS model outputs of +1 to -1 are arranged in
7 class system to facilitate comparison and communication.
Score range
Class
+1.0
Fully suitable or highest suitability
0.99 to 0.50
Moderately suitable
0,49 to 0.01
Somewhat suitable
0.0
Uncertainty
-0.01 -0.49
Somewhat unsuitable
-0.50 -0.99
Moderately unsuitable
-1.0
Fully unsuitable or lowest suitability
a
61
APPENDIX 2.
Example Questionnaire provided to panelists
EXAMPLE QUESTIONNAiRE
Thank You
Thank you for agreeing to serve as an expert panelist for this project. I have a
because of your knowledge about coho salmon Oncorhynchus kisutch and their relat
freshwater habitat conditions. I realize that you have other responsibilities and priorities and therefore I
value any time and input you can lend to this project. Your participation and ideas shared in this
exercise will remain anonymous. I hope you are able to continue participating on this panel. Thank you
again for your time and efforts. I look forward to receipt of your input.
Instructions
The purpose of this exercise is to create a decision-support model (DSM) to assess the ecological
condition7 of watersheds for coho salmon within Oregon's Coastal Coho ESU (coastal watersheds
between the mouth of the Columbia River and Cape Blanco). This expert panel exercise is necessary
because field data and literature are currently inadequate for developing such a model. The goal is to
come to agreement on the structure of the proposed model, what parameters should be included in the
model, and how the model should operate. This is the 1 of probably 4 rounds in this exercise.
Background information is provided at the end of this document to explain the model.
Below each question are specific instructions. Please consider all of the provided information when
answering each question. There is room for your comments, ideas, logic, references, etc. As best you
can, please include reasons for your answers. It is important that you use your "gut" feeling or opinion,
even if no data are available, If you do mention a reference, please include enough information for me
to find the source. Please rate the confidence of your answers on a scale of 1-5 (i.e. 1= gut feeling, 5'
extensive documentation). You may also want to draw or create diagrams to share your ideas. For that I
am mailing you a paper copy of this questionnaire with a self addressed stamped envelope for its return.
Also, feel free to share this questionnaire and work through its questions with your co-workers and
other colleagues.
If you have any questions don't hesitate to call or email me. Please return your response within 10 days.
I will receive your returned comments, summarize all of the participants' input, and send you a new
questionnaire. This process will continue until the panel of experts has come to agreement.
a watershed to be ecologically intact it means: (1) the processes necessary to create and maintain
habitat conditions for native species and especially coho are intact; (2) current habitat conditions
support coho; and (3) the system has the potential to recover to desired conditions when disturbed by
large natural events or by land management activities.
62
Model Use
The Aquatic and Riparian Effectiveness Monitoring Plan (AREMP) for the Northwest Forest Plan uses
Decision support models to characterize ecological condition of watersheds and aquatic ecosystems. To
do this, the AREMP assessment team uses a decision support modeling tool called Ecosystem
Management Decision Support (EMDS), which consists of GIS, database, and decision-support
software components. EMDS enables AREMP users to evaluate monitoring data from instream,
riparian, and upslope conditions, and then aggregate the information into an overall score of watershed
condition. Experts were consulted in an informal group process to create AREMP's model. This
involved deciding which parameters to use and then creating the rules for how to evaluate and
aggregate them.
I am building a coho-centric DSM to assess watershed condition for Oregon's coastal coho salmon ESU
that is based on AREMP's coastal watershed model. I will use both the same modeling program,
structure, and rules for arranging and aggregating watershed parameters. The only major difference
between the two models is the parameters. The parameters I have chosen all come from the State's
assessment of the coastal coho salmon ESU. Criteria to evaluate each of the parameters will be provided
by the state agencies that manage the data sets. These criteria for evaluating the data are relevant to
coho salmon, are specific to the ESU, and were used in Oregon's coastal coho salmon assessment.
Model Basics and Structure
A decision-support model is simply a method of documenting a formal, logical organization of
information for evaluation and interpretation. This allows for the decision process to be applied
consistently and transparently.
A simplified diagram of the structure and function of the model is shown in Figure 1. This decisionsupport model has two primary functions. First, it evaluates individual data and secondly, it aggregates
this information into an over all assessment of condition.
Figure L
Combined Score
I
I
I
Aggregate
Evaluate / Normalize
Watershed Parameters
(data)
For each parameter used in the model, experts must
develop criteria to evaluate the data. The approach is
similar to developing habitat suitability curves and
defining the benchmarks or break points for good and
poor condition. Data is then evaluated and normalized
based on this criterion and given a score between +1 to -1.
A score of+l indicates that the parameter is in good
condition or is fully suitable to sustain healthy runs of
wild salmon. Conversely, a score of -1 indicates the
parameter is fully unsuitable. Evaluation scores between
+1 and -1 reflect the gradient between good and poor
conditions. Converting scores to this simple scale of +1 to
- I facilitates comparing and aggregating scores later in
the model.
After evaluation, individual parameter scores are
aggregated into an overall score of condition in multiple,
sequential steps in a model structure that resembles a tree
diagram, as illustrated by the diagram of the proposed
model (Figure 2). The model structure is important
because it determines the logical arrangement of parameters. At each junction where two or more
63
parameters are aggregated, expert defined rules determine whether the minimum score, maximum
score, or average score will be passed on to the subsequent level. Experts may also chose to weight
certain parameters for added influence at each junction where they are combined. Parameters are
aggregated logically until all the parameters in the model have been combined. The final watershed
condition score is also a number between +1 and -1, reflecting its condition relative to coho suitability.
From the top down you'll notice 4 colored branches of the proposed model structure (Figure 2). The
top two are vegetation (green) and roads (brown), which are combined to form the watershed condition
drivers category. All of the parameters in the drivers category directly influence watershed condition.
The bottom two branches are water quality (blue) and reach condition (orange), which are combined to
form the responses category. All the parameters in this category respond to the drivers. The reach
condition branch (orange) is aggregates biological and physical conditions.
At each junction where two or more parameters are aggregated there are small boxes
containing the words "AVE" or "MIIN." These are called operators and they determine how antecedent
parameters are combined. MTIN operators pass a score weighted towards the lowest evaluation score.
MIN is used primarily to allow one indicator to override other indicators. AVE operators pass on the
average evaluation score. AVE is used so that indicators in good and poor condition balance each other
out. An operator that is not on the current model but can also be used is MAX. MAX passes the highest
evaluation score and therefore presents an optimistic view of condition.
Questions
Question 1: Model Structure
The main purpose of this exercise is to help develop and evaluate the model structure using expert
opinion. Consider the structure of the model and the hierarchical arrangement of the parameters. What
comments, suggestions, improvements, or concerns do you have? Would you change the structure in
any way? What levels would you add or subtract?
Question 2: Parameters
The parameters I have chosen for my model must meet the following criteria: (I) They must act as
surrogates or indicators of watershed processes and conditions that influence coho; (2) Data for the
parameters much currently exist (in data bases or GIS layers) and be available to the public; (3) Data for
the parameters must exist for the entire region encompassed by the coastal coho ESU. All of the
parameters I selected for my model were used in Oregon's coastal coho ESU assessment. I have
included a list of these parameters and their definitions at the end of this document.
Do you know of any other parameters that I have not included in the model that meet these criteria?
Please list them, and include their source, rational for their use, and describe where they belong in the
model structure.
How certain are you about your decision? (1=No certainty, 5"Very certain and have the literature to
prove it.)
1
2
3
4
5
64
Question 3: Operators
The operators AVE (average), MIN (minimum), and MAX (maximum) determine how two or more
parameters' scores are aggregated at each junction.
Examine the operators at each junction in the model. From your limited knowledge about the structure
of this model and the role of operators, do you agree with the operators I have chosen? What changes,
concerns, recommendations, or thoughts do you have?
How certain are you about your decision? (1=No certainty, 5=Very certain and have the literature to
prove it.)
1
2
3
4
5
Question 4: Weighing Parameters
As of now, all of the individual parameters are weighted the same.
Would you weight any of these parameters differently than others? Which ones? Please give your
rationale, comments and suggestions.
How certain are you about your decision? (1=No certainty, 5=Very certain and have the literature to
prove it.)
1
2
3
4
5
65
Figure 2. The proposed model structure from AREMP. The structure, categories, and operators are all
consistent with AREMP's coastal watershed condition model.
Vegetation
Condition of vegetanon
in watershed
Drtvefs
Condition o' vatiablen
thet drive or directly
influence watershed
condition
<AVE
Rosda
Condiflon of roads in
watershed
(
I
WATERSHED
CONDITION
Conctflo,rs are suitable
to sustain urable rurs of
ld cohn salmon
(,A
Water Quailty
Evaluation of eater
quality parameters
/
Phyulcat Condition
Responses
Condition of vonshies
thot respond to drfuinq
influences
Evaluation of reach
physical condition
autibutes
<AVE
/
R.IcI, Condition
Aoerage of reach
condition scores
AVE operators pass the average evaluation score
BiologIcal Condition
Evslustion of react,
bislolcal conditioO
attributes
4.
I
4.
66
Table 1. Parameters used in Oregon5s assessment of the Oregon Coastal Coho Salmon
ESU
Parameter
Percent of watershed area m urban or agncultural land use
Percent of Watershed in Urb/Arg
Percent of 180 degree sky that is shaded by trees or other topographic
features
The number of conifers > 50cm dbh within 30 m of both sides of the
river per 305 m of primary stream length
% Shade
Conifers > 50cm dbh in Riparian
The number of conifers >90cm dbh within 30 m of both sides of the
river per 305 m of primary stream length
State rating of 1 -yen to roads based on cuffent condition including:
erosion, washout potential, culverts limiting fish passage, and
drainage
Density of roads with SOm of the stream channel
Conifers> 90cm dbh
Road Condition Index
Roads in
Density of roads on slopes > 50%
Roads on steep slopes
Number of stream crossing culverts per kilometer
stream crossing eulverts
Phosphorous
Evaluation of total phosphorous concentration (rng/L) in the watershed
Nitrogen
Evaluation of the total inorganic nitrogen concentration in the watershed
Hydrogen ion activity
pH
Evaluation of the average dissolved oxygen collected in the field
surveys.
Evaluation of the 7-day average maximum water temperature.
Dissolved Oxygen
Water Temperature
% fine sediment
Fine Sediment
Total solids concentration in water column as indication of suspended
and dissolved component of sediment.
Visual estimate of stream substrate composed of solid bedrock
Total solids
% Bedrock
Visual estimate of stream substrate composed of 2-64mm diameter
particle
Visual estimate of substrate composed of <2mm diameter
% Gravel in Riffles
% Fine Sediment in Riffles
% of primary channel area represented by slackwater pool habitat
(beaver ponds, backwater, alcoves, and isolated pools).
% of primary channel area, represented by pool habitat
% Slackwater Pools
% Pools
Pools> lmdeepperkmofprimamy channel
DeepPools/km
% total channel area represented by secondary channels.
% Secondary Channel
Pieces Large Wood
Key Pieces Large Wood
Large Wood Volume
Vertebrate Community Score
Macroinvertebrate Community Score
Exotic Fish & Amphibians
-
4 of pieces of wood >0.15 in diameter x 3m length per 100 m primary
stream length
4 ofpieees of wood 60cm diameter & 12 m long per 100 m primary
stream leng
Volume ofwood>0.15 mx3 mlengthper lOOmeters primary stream
length.
An index of biotic condition based on species diversity that was created
for coldwater streams in Western Oregon and Washington
River Invertebrate Predication and Classification System, or RJVPACS.
An index of biotic condition based on species diversity.
Download